Mailman 3 April 2021 - DIPY

Open PhD Position on Machine Learning for Brain Connectivity in Clinical Neuroscience
by Emanuele Olivetti April 28, 2021

April 28, 2021

Dear DiPy community, We are pleased to announce the opening of one PhD position in neuroinformatics at the International Doctoral School in Cognitive and Brain Sciences (CIMeC) of the University of Trento, Italy. The PhD Grant is funded by the Neuroinformatics Laboratory (NILab) of Fondazione Bruno Kessler (FBK). The position is part of the “Neusurplan” project, an integrated approach to neurosurgery planning based on multimodal and longitudinal data. Neuroimaging methods, like structural, functional, and diffusion magnetic resonance imaging (MRI), can be used to investigate the anatomical and functional connectivity of the brain. In this project, the candidate will pursue research on machine learning methods for neuroimaging data analysis to study and characterize brain connectivity, with applications to longitudinal studies and clinical practice. The ideal candidate should have a mixed background in neuroimaging techniques and numerate disciplines, like computer science, engineering, physics, or mathematics. This project is in collaboration with the Division of Neurosurgery, S. Chiara Hospital, Trento (IT). The position is for a 4 year PhD program (Nov. 1, 2020- Oct. 31, 2024). Courses are in English. The salary is starting at approximately €1.200/mo., net. An additional personal budget of around €5.000 is provided for research and mobility. The University of Trento ranks among top Italian Universities ( https://www.unitn.it/en/ateneo/1636/rankings). Fondazione Bruno Kessler ranks first among the Italian research centers in Engineering and Computer Science ( https://magazine.fbk.eu/en/news/fbk-ranks-1st-in-italy-for-scientific-excel… ). Why choose Trento? ( http://www.unitn.it/en/ateneo/1629/why-choose-the-university-of-trento) Important dates: Deadline for application: 25 May 2021 Evaluation interview: 28 June 2021 Beginning of PhD program: 1 November 2021 Apply now: https://www.unitn.it/en/ateneo/1942/announcement-of-selection https://www.unitn.it/drcimec/105/2021-topic-specific-grants-and-descriptions You are kindly invited to contact in advance Emanuele Olivetti ( olivetti(a)fbk.eu) and Paolo Avesani (paolo.avesani(a)unitn.it). Recent related publications: - Giulia Bertò, Daniel Bullock, Pietro Astolfi, Soichi Hayashi, Luca Zigiotto, Luciano Annicchiarico, Francesco Corsini, Alessandro De Benedictis, Silvio Sarubbo, Franco Pestilli, Paolo Avesani, Emanuele Olivetti, "Classifyber, a robust streamline-based linear classifier for white matter bundle segmentation", Neuroimage, 2021, https://doi.org/10.1016/j.neuroimage.2020.117402 - Gabriele Amorosino, Denis Peruzzo, Pietro Astolfi, Daniela Redaelli, Paolo Avesani, Filippo Arrigoni, Emanuele Olivetti, "Automatic Tissue Segmentation with Deep Learning in Patients with Congenital or Acquired Distortion of Brain Anatomy", Machine Learning in Clinical Neuroimaging and Radiogenomics in Neuro-oncology (MLCN 2020), https://doi.org/10.1007/978-3-030-66843-3_2 - Pietro Astolfi, Ruben Verhagen, Laurent Petit, Emanuele Olivetti, Jonathan Masci, Davide Boscaini, Paolo Avesani, "Tractogram filtering of anatomically non-plausible fibers with geometric deep learning", Medical Image Computing and Computer Assisted Intervention (MICCAI 2020), https://doi.org/10.1007/978-3-030-59728-3_29 -- Emanuele Olivetti, Ph.D. Senior Research Scientist at Bruno Kessler Foundation (FBK-ICT) NeuroInformatics Laboratory (NILab) http://nilab.fbk.eu Center for Mind and Brain Sciences (CIMeC), University of Trento Via delle Regole 101 - 38123 Mattarello (Trento), ITALY olivetti(a)fbk.eu - +39 0461314179 emanuele.olivetti(a)unitn.it - +39 0461282760 -- -- Le informazioni contenute nella presente comunicazione sono di natura privata e come tali sono da considerarsi riservate ed indirizzate esclusivamente ai destinatari indicati e per le finalità strettamente legate al relativo contenuto. Se avete ricevuto questo messaggio per errore, vi preghiamo di eliminarlo e di inviare una comunicazione all’indirizzo e-mail del mittente. -- The information transmitted is intended only for the person or entity to which it is addressed and may contain confidential and/or privileged material. If you received this in error, please contact the sender and delete the material.

1 0

AI / Machine Learning developer (full-time position)
by Benoit Scherrer April 26, 2021

April 26, 2021

Hi Dipy community, We are hiring at Quantivly - I can see a lot of common interest & expertise, so I thought I would share. *Description* Our goal at Quantivly is to empower Radiology to make data-driven decisions. Today, MRI scanners in the US are idle, not taking images more than 60% of the day, wasting $10B+ in potential revenues annually and delaying patient care. Unfortunately, improving the process of acquiring medical images is widely overlooked, largely because the data needed to drive productivity is simply inaccessible. Quantivly, a spin-off from Boston Children’s Hospital, has developed a platform that analyzes MRI images metadata on-the-fly to unlock operational data and radically transform the way MRI operation is managed. We are looking for a AI developer who can help take our startup to the next level. You will be the lead AI developer. You will be expected to take end-to-end ownership of key ML/AI features on our roadmap and execute to deliver solutions that revolutionize imaging operations in radiology and delight customers. We are passionate about making a difference in health care by bringing data science tools to Radiology. This is your chance to build your career at a company that is at the forefront of changing its industry. We need brains, ambition, and passion to continue driving this change — join Quantivly! *Compensation* Salary: $100-120k, Equity: 0.25%-1.0% *Education* Ph.D. in computer science or similar practical experience in an industrial or research setting. *Responsibilities* - Assist with data collection efforts, de-identification and data lake construction. - Design, implement and evaluate machine learning and deep learning solutions on 3-D/4-D image data and/or image metadata - Conduct machine learning tests and experiments, understand results to refine models and/or pre-processing to optimize performance - Deploy solutions for production on the Quantivly platform - Publish and disseminate methods and findings - Keep up-to-date with the latest advancements in ML/AI *Required* - 3+ years’ experience designing, developing, and implementing ML / AI algorithms - Strong programming skills in python, including fundamental software engineering principles and machine learning design patterns. - Experience with at least one Deep Learning framework such as TensorFlow, PyTorch, or Keras, or ML framework such as Scikit-learn - Demonstrated understanding in AI/ML algorithms, learning models, graphical models, overfitting, regularization, etc. - Experience in building ML pipelines for processing large datasets. - Understanding of state-of-the-art approaches to computer vision - Refereed publications in premiere computer vision and/or bioinformatics venues. - Deep knowledge of math (linear algebra), probability, statistics and algorithms - Outstanding analytical and problem-solving skills - Work well with in a fast moving, small team environment *Who you are* - You are highly motivated and have a passion for creating and supporting great products. - You're curious, love to learn and to dig into new technologies, and can pick them up quickly - You're a go-getter and seek ownership of projects with the demonstrated ability to deliver results. - You thrive on collaboration, working side by side with people of all backgrounds and disciplines, and you have very strong verbal and written communication skills. - You are great at solving problems, debugging, troubleshooting, designing, and implementing solutions to complex technical issues. - You strive to excel, innovate and take pride in your work *Bonus qualifications* - Experience in working with medical images - Experience with the MONAI framework - Knowledge of DICOM and HL7 standards - Solid experience in Linux/Unix, shell scripting, git, docker. - Knowledge of SQL. Experience with SQL server-side machine learning (eg PL/Python) - Knowledge of GANs and ML techniques for generating synthetic data, NLP, ... *Our tech stack* - Python3, Java, C/C++, Docker - Postgres 13, GraphQL, Flyway, Redis, Elasticsearch, - Slurm, dcm4che, Websocket, WAMP/crossbar, ... *Working locations & additional information* - Somerville, MA (remote option) - Must be authorized to work in the United States - Travel: Infrequent - Employment Status: Full-Time Employee with benefits (Medical, Dental, Vision) *Equal Opportunity/Affirmative Action Statement* At Quantivly we celebrate difference. We are committed to ensuring an environment of mutual respect for every employee and proud to be an equal employment opportunity employer who does not discriminate against any person because of race, color, creed, religion, gender, gender identity, gender expression, national origin, citizenship, age, sex, sexual orientation, pregnancy, marital status, ancestry, physical or mental disability, military or veteran status, or any other characteristic protected by law. We believe a diverse and inclusive workplace is central to our success and actively seek to recruit, develop and retain the most talented people from a diverse pool of candidates. *Contact:* jobs(a)quantivly.com Keep up the great work @Dipy Best Benoit -- Benoit Scherrer CEO @ Quantivly benoit(a)quantivly.com

1 0

Strange MD and FA values in preclinical rodent data
by jarcos＠iib.uam.es April 23, 2021

April 23, 2021

Greetings DIPY experts, I'm Javier Arcos Hódar, a research assistant based on the Alberto Sols Institute for Biomedical Research on Madrid. I've been tasked to investigate other software for MRI data processing in case we end up having problems with our homemade Matlab software. First of all I'd like to be thankful for the developing of DIPY, since it seems to work faster, is fully open with helpful tutorials online and provide options our homemade software doesn't. Unfortunately I might be mismanaging something since even for the very simple task I wanted to start with (exporting AD, MD and FA maps) it's giving me some problems. I'll offer my files and anaconda spyder workspace in case it is helpful. We use 2 basal images, 2 b values beyond basal (~300-~1200) and 7 directions: https://www.mediafire.com/file/512di21tmeb9av3/javierdipymailquestion.rar/f… Note b table and directions are directly in the workspace; I got annoyed trying to import them as the tutorial says due to an error of not formatting the text file properly and created a matrix of the same dimensions and entered the values manually as python objects. To clarify, the files from our alternative software are not a .nii stack but can be opened on imageJ under (checking CorteX folder, not the CorteX Config folder) Import -> Raw -> DTI_FA_Raw, with settings (I think those are the default but please double check to avoid problems) -image type 64-bit Real -Width and Height 128 pixels, -offset of first image 0, -number of images 1, -gap between images 0 bytes, and -only little-endian byte order ticked on, with the three other settings on blank. Inferred eigenvalues and eigenvectors can be checked in the DTI_Data txt file for each 'corte' (slice). I picked two example subjects were nearly every voxel had a curve fit above >75 just to try and make certain noise/phantoms Three issues are obvious: -MD values for each voxel appear a million-fold lower than in our Raw files but otherwise quite closely aligned. The most obvious, but perhaps wrong, explanation is that they are given in cm²/s instead of µm²/s as the output of our other software. This is a bit strange to me, as in: is some information missing from the way I imported the data that SI units are used, or is clinical human data typically given in cm²/s? It's not hard to solve this, whether in fslmaths or probably within dipy, but still raises my concern. -Fractional anisotropy values are more problematic. Taking a rough whole-brain ROI in ImageJ and dropping NaN values in to compare first slice of rat 19 in R, the difference averages .0837, which is big compared to the actual values and their variability. Moreover, it seems mostly driven by values that ought to be high, in ventricles or the dead core of a brain tumour (.30 or greater) instead becoming wildly high (>.50 or >.60) to the point it probably isn't accurate or biologically plausible. See for example Li XY et al (2017) https://www.sciencedirect.com/science/article/pii/S1995764517304492. I completely ignore the cause for this; I might be misunderstanding something but if it's a variance, multiplying everything by the same number shouldn't drive big differences in values, so the previously mentioned MD thing shouldn't be driving this. I get a >.8 correlation between difference between software output and dipy value, so I'm pretty sure still the problem really is curve fitting in high values, but I'm at loss at how this would be solved, something like scipy.optimize.curve_fit? -Checking the manual, (https://dipy.org/documentation/1.4.0./reference/dipy.reconst/) it seems ADC values are only offered voxel by voxel, without an obvious way to run it over the whole image. This causes me a bit of trouble comparing them for each direction to try and find the cause of the problem. It wouldn't be too hard to make it iterate over all voxels but I'd like to know if there's an inbuilt way to do this instead. We use two basal images per slice, and I understand that just having 0 b values would make DIPY interpret them as this, but perhaps the way to handle them (beyond simply averaging them?) could also cause differences? I don't know, but since similar issues have happened trying other software (dsi studio, track vis) I was hoping to check if in this more open software with community support the issue can be diagnosed/corrected. Lastly, note that I didn't use mask in the workspace provided since for our resolution and number of directions processing isn't too computionally demanding and the curve-fitting is done voxel by voxel anyway, but masks are provided in the folder in case someone wants to use them (it would remove the ugly background in the final nifty files at least) Thanks in advance

1 0

Postdoc position in diffusion imaging
by Cieslak, Matthew April 21, 2021

April 21, 2021

Hi Dipy community! Our group is looking to hire a postdoctoral fellow to work with us on diffusion imaging. Specifically we’re planning on expanding QSIPrep’s reconstruction workflows to use Dipy, MRtrix3 and DSI Studio’s group analysis tools, with application to large developmental datasets. The job listing is below, please contact us if you have any questions and feel free to share the listing with anyone who might be looking for a postdoc job. Description Ted Satterthwaite, Matt Cieslak, and the Penn Lifespan Informatics and Neuroimaging Center (PennLINC, www.pennlinc.io<http://www.pennlinc.io/>) is seeking a post-doctoral fellow to develop advanced methods for diffusion-weighted imaging, with applications to studies of the developing brain. Specifically, the fellow will build on the success of QSIPrep (qsiprep.readthedocs.io) expand the repertoire of advanced methods included in this highly reproducible and scalable framework; further integration with the Dipy ecosystem is of particular importance. These methods would then be applied to large-scale studies of brain development including the Philadelphia Neurodevelopmental Cohort (n>1,600), the Healthy Brain Network (n >2,500), and ABCD (n>11,000) to understand both normal white matter development and how abnormalities underlie neuropsychiatric syndromes. As such, the position will involve working as part of a highly interdisciplinary research team, with particular emphasis on collaborative work with leaders including Matthew Cieslak (creator of QSIPrep), Elef Garyfallidis (Indiana U; creator of Dipy), Ariel Rokum (UWash; neuroinformatics), Damien Fair (UMinn; ABCD imaging), Danielle Bassett (network science), Taki Shinohara (imaging statistics), and Mike Milham (Child Mind Institute; PI of HBN). To work effectively in this highly collaborative environment, the applicant must have superior communication, language, and writing skills. Salary is quite flexible and based upon experience; position is contingent upon funding. Applicants should contact Ted Satterthwaite (sattertt at upenn dot edu). Requirements The applicant must have completed their Ph.D. in, engineering, computer science or computational neuroscience (preferred), psychology, or statistics with an established record of high productivity. Expertise in Python, R, and neuroimaging software (e.g., ANTs, MRtrix3, FSL diffusion tools) is required; specific experience with diffusion imaging, Dipy, NiPype, and Docker is highly desirable. • Location: Philadelphia, Pennsylvania, United States • Position Title: Post-doctoral fellow: Diffusion Imaging • Company Name: University of Pennsylvania • Job Function: Engineering • Job Type: Full-Time • Job Duration: 1-2 Years • Min Education: Ph.D. • Min Experience: 3-5 Years • Required Travel: 0-10%

1 0

Reminder: DIPY Open Meeting - Q-Space Trajectory Imaging
by Eleftherios Garyfallidis April 15, 2021

April 15, 2021

Hello all, Dr. Leevi Kerkelä (UCL) will discuss tomorrow his project on integrating Q-Space Trajectory Imaging in DIPY. The talk is tomorrow Thursday 15 April at 1pm EST. The zoom link for the open meeting is : https://iu.zoom.us/j/84926066336. Best, Eleftherios p.s. More talks for the bi-weekly open meetings will be soon announced here https://github.com/dipy/dipy/issues/2312

1 0

patch2self denoising attenuated b=0 images
by lmuftuler＠mcw.edu April 1, 2021

April 1, 2021

Hi, I was trying out the new patch2self denoising on a 4-shell (b=1k s/mm^2 up to 4k s/mm^2) diffusion data (total of 94 directions plus 4 b=0 images). First, I noticed that it attenuated the b=0 images (all four of them) with respect to the rest of the image series. This in turn reduced map-mri ZD maps by about 10% to 20%, which was expected b/c (E(q)) essentially got "smaller". Did anybody have any experience with this? Any suggestion why it happened and how to fix it? best, -Tugan

4 4